Artificial Intelligence
95% of companies surveyed by Bain already use AI. What can small and medium businesses learn?
Artificial intelligence is no longer a distant possibility for Brazilian companies. The debate no longer starts only with the question of whether or not to use the technology. The more relevant question has become another: in which processes is AI entering, what problem is it solving, and how much value can it return to the business?
This adoption journey connects directly to my research path. In the Professional Master’s in Economics at Fundação Getulio Vargas, I developed the IBEOP-IA, Brazilian Index of Occupational Potential Exposure to Artificial Intelligence, to investigate how AI capabilities can reach different Brazilian occupations. The study offers a perspective on the potential for incorporating the technology into tasks and work. The market research presented in this article complements that analysis by showing how adoption is already happening in companies and in which processes AI is beginning to produce concrete results.
2026 data show that AI has already reached small businesses, but the next challenge is turning tools and tests into processes that generate measurable results.
This shift appears in the survey AI use reaches 95% of companies in Brazil, published by Bain & Company in 2026. The research heard 140 Brazilian companies, with annual revenue between R$ 40 million and more than R$ 10 billion, in sectors such as financial services, retail, technology, telecommunications, industry, health, energy, and consumer goods.
Among the organizations surveyed, 95% were already using artificial intelligence. In 2024, the percentage was 74%. The 21 percentage point advance in two years shows that access to the technology spread quickly in this group.
The figure is significant, but it needs to be interpreted within its scope. The research does not automatically represent all Brazilian companies. It portrays a set of medium and large organizations that already have structure, budget, and competitive pressure to advance adoption.
Still, the survey offers a relevant message for companies of different sizes: when technology becomes accessible to many competitors, simple access stops being the main source of differentiation.
Adoption grew, but scale still separates companies
The 95% indicate presence of AI, not necessarily operational maturity.
According to Bain, about 40% of surveyed companies have already scaled at least one use case. That means the majority still lives with point initiatives, experiments, or applications restricted to certain teams.
At the same time, 62% report direct financial impact and 91% say their initiatives met or exceeded expectations. For the next cycle, 98% intend to expand the budget allocated to the technology.
These data show growing confidence, but they do not eliminate the execution challenge. The distance between experimenting and operating remains relevant. An individually used assistant, an automation restricted to one team, and an agent connected to corporate systems represent very different levels of complexity, responsibility, and risk.
The survey itself points out that AI is already among the five main strategic priorities of 86% of surveyed companies. In 62% of cases, programs are led by the CEO or by executives who report directly to them.
When the initiative reaches that level, it ceases to be only a technology project. It becomes a decision about processes, data, people, controls, and operating model.
Growth happened where work happens every day
AI use advanced mainly in areas that keep the company running continuously.
In the comparison presented by Bain, IT led growth, with a 25 percentage point increase. Operations grew 23 points and customer service, 16 points. Support for analytical and intellectual work, such as preparing reports, research, and data analysis, advanced 14 points.
These areas combine characteristics favorable to AI application. They deal with large volumes of information, recurring activities, context-based decisions, and a constant need to record, classify, respond to, or route requests.
They are also areas in which small improvements can repeat many times. Reducing a few minutes from a rare activity generates little impact. Reducing the time of a task executed thousands of times can change cost, capacity, and customer experience.
The survey also records an 11 percentage point drop in the use of traditional chatbots. The research does not explain the cause. Therefore, it would not be correct to conclude that those solutions ceased to be useful. One possible interpretation is that part of companies may be replacing isolated interfaces with applications more integrated into processes, but that hypothesis would require additional evidence.
The data reinforces an important caution: measuring only the presence of a tool says little about the value it produces.
AI has reached small companies, but still needs to enter processes
A smaller company does not have the same scale, budget, or structure as the organizations surveyed by Bain. That does not mean it is outside the adoption movement.
A Sebrae survey published in November 2026 shows that, among small businesses consulted about marketing and communication, 29% were already using AI tools regularly and 21% had run tests. Another 30% planned to use the technology. In that specific scope, half of respondents had already used or experimented with AI, and eight in ten combined use, tests, or intention to adopt.
Another 2026 survey, conducted by Strand Partners for AWS with a representative sample of a thousand business leaders in Brazil, indicates that 50% of companies were already using AI. Among adopters, 58% remained in basic applications, 27% were at an intermediate stage, and 15% had reached advanced uses. Among small and medium businesses, 13% reached that advanced stage.
The Bain, Sebrae, and Strand Partners surveys use different samples, questions, and definitions. Their percentages should not be compared as if they measured exactly the same universe. Together, however, they show a consistent direction: AI has already reached companies of different sizes, but the depth of use still varies considerably.
For small and medium businesses, the challenge is not to reproduce the structure of large organizations. It is to choose a relevant process, understand its current cost, and advance with an application compatible with the business’s capacity to invest, manage, and monitor.
Many of the processes described by Bain appear in companies of different sizes. Four starting points are especially easy to recognize:
- Customer service: classify requests, prepare responses, consult information, and route exceptions to a person.
- Operations: check registrations, documents, billing, orders, or incidents that follow known rules.
- Reporting: gather recurring data, produce a first analysis, and prepare information for meetings and decisions.
- Commercial: track proposals, organize interactions, remember next steps, and prepare information for customer contact.
These examples do not mean that every repetitive activity should receive AI. Processes with simple, stable rules may be handled by conventional automation. Others require judgment, language processing, consultation of different sources, or adaptation to context and may justify the use of AI models or agents.
The choice should start from the problem, not from technological novelty.
The first step is to measure the current process
Before introducing AI, the company needs to understand how work happens today.
How much time does the activity consume? How often does it occur? What is the error or rework rate? Which steps depend on approval? What information is used? What happens when the case falls outside the pattern?
Without a baseline, any gain can seem convincing. A quick demonstration can impress, but it does not inform whether the solution is reliable, economically viable, or suitable for recurring use.
A practical path can be organized in five moves:
- Choose a delimited, recurring, and relevant process.
- Record the cost, time, quality, and exceptions of the current process.
- Define what AI may do, which data it may use, and when it should request human review.
- Test with representative cases and track result, failures, consumption, and rework.
- Expand only when there is evidence of benefit and operational conditions to sustain use.
The goal is not only to execute the same task with a new tool. In many cases, it will be necessary to review steps, responsibilities, and decision points.
The global survey The state of AI in 2026, from McKinsey, helps size this challenge. Although AI use continues to grow, 37% of respondents attributed some positive impact on EBIT and only 6% were classified as high-performing organizations in AI. According to the study, these organizations tend to redesign workflows, measure impact, and combine leadership, risk management, and operational discipline.
The Bain and McKinsey percentages are not directly comparable, because they use different samples, questions, and scopes. Together, however, they help show that broad adoption does not guarantee broad results.
Value needs to appear in the complete process
A local improvement may only shift the bottleneck.
If AI prepares a response more quickly, but human review remains backlogged, total time may not fall. If an agent classifies requests but routes many cases to the wrong queue, apparent productivity may generate rework. If a report is produced in seconds but contains inconsistent information, the cost of correction may exceed the initial savings.
Therefore, measurement needs to consider the complete process. Depending on the case, indicators may include cycle time, completion rate, output quality, number of exceptions, rework, cost per task, need for review, and effect on the customer or on operations.
It is also important to record when AI should not act. Processes involving sensitive decisions, protected data, financial commitments, or changes in critical systems may require human approval, limits on autonomy, and additional controls.
When AI enters the process, an operations problem begins
While the technology only suggests text for a person, risk remains relatively bounded. When an agent consults systems, selects tools, updates records, sends messages, or triggers other actions, the company begins to delegate part of execution.
At that point, operational questions arise:
Who is responsible for the agent? Which version is in use? Which systems and data can it access? How are its actions recorded? How are failures, repetitions, or consumption outside expectations identified? Who can stop execution? How is it verified that the task was actually completed?
The NIST AI Risk Management Framework organizes AI risk management into four functions: govern, map, measure, and manage. The model reinforces that assessment should not happen only before deployment. AI systems need to be tested and monitored during operation [5].
This need is especially important for agents. Because they can combine models, data, tools, and decisions in the same trajectory, evaluating only the final response can hide intermediate failures.
In this context, AgentOps becomes relevant. The discipline seeks to apply operation, observability, assessment, control, and continuous improvement practices to AI agents in production.
From process to operation: ARGUS
ARGUS acts as an AgentOps platform to support monitoring, orchestration, and governance of agents integrated into the environment.
The platform can centralize information about owners, versions, models, tools, dependencies, and operational context. It can also track executions, actions, failures, consumption, and indicators defined for each implementation.
This visibility helps relate agent behavior to the process it was supposed to execute. Instead of observing only model calls or isolated messages, the organization can track whether the task advanced, where a failure occurred, how much resource was used, and when there was human intervention.
ARGUS does not on its own choose which processes should receive AI, does not replace organizational diagnosis, and does not define the company’s business rules. Its coverage depends on agents, models, tools, and telemetry sources actually integrated. The platform adds supervision and operational evidence after objectives, responsibilities, limits, and integrations have been defined.
About Target Solutions
Target Solutions has spent more than 15 years supporting organizations in evolving their infrastructure, monitoring, observability, automation, and operational intelligence operations.
Throughout this journey, it has participated in projects in complex environments in Telecom, Technology, Government, Energy, and large companies, contributing to increase operational efficiency and service reliability.
With the advance of AI agents, this experience extends to implementing AgentOps practices, governance, supervision, and operational control.
We believe AI generates sustainable value when it enters well-understood processes, with measurable objectives, defined responsibilities, and capacity for monitoring throughout operation.
Beyond the platform, Target Solutions offers specialized services to support adoption of AI agents, from diagnosis and inventory of initiatives to defining governance, processes, architecture, integrations, observability, and operating model. This work helps prepare the organizational and technological environment needed to implement and use a platform such as ARGUS in a structured way.
If your company wants to identify a first process to apply AI, assess initiative viability, or structure agents with governance and operational monitoring, learn about Target Solutions services for AI agents and talk with us about how to advance in a way compatible with business reality.
About the author
Paulo Florêncio Santos de Menezes is partner and co-founder of Target Solutions, with work in infrastructure, monitoring, observability, automation, IT operations, and AI agent governance.
He holds a degree in Data Processing from Universidade Federal do Amazonas, an MBA in Business Management from Fundação Dom Cabral, and a Professional Master’s in Economics from Fundação Getulio Vargas.
In his Master’s research, he developed IBEOP-IA (Brazilian Index of Occupational Potential Exposure to Artificial Intelligence), investigating the potential exposure of Brazilian occupations to AI capabilities through the combination of occupational data, artificial intelligence patents, and natural language processing techniques.
References
[1] Bain & Company. AI use reaches 95% of companies in Brazil. 2026.
[2] Sebrae. Eight in ten small businesses already use or plan to use AI for marketing and communication. 2026.
[3] Strand Partners. Unlocking AI’s potential in Brazil. Survey conducted for AWS, 2026.
[4] McKinsey & Company. The state of AI in 2026: On the road to ROI. 2026.
[5] National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework (AI RMF 1.0). 2023.